Optimization has got to be the most overused word when it comes to my line of work. If I had to take a guess I would say no more than 1 in 5 optimization claims are anything more than a design study. It seems to have come to the point that if a person compares two analyses, and one is better, the claim is that it has been optimized. As someone who scopes the work required for design services I have some thoughts about “optimization”.
When a customer approaches me to scope work for either an electronics cooling, clean room, data center, etc. application there are three general classifications that I use to break down the amount of work involved.
- Analysis: In this scenario the customer essentially owns the design and the cost is associated with each analysis and report. This works well when there are significant constraints unrelated to thermal/airflow that will drive the design. With our recommendations they can weigh all of the constraints and opportunities and decide in what direction the design should proceed.
- Design: In this scenario the customer would provide the constraints but we own the design and the number of analyses required to reach the goals. As an example, their constraints might be: Off-the-shelf heat sinks only, maximum height of a heat sink, component locations, maximum operating temperatures, etc. Once the constraints have been defined the customer pays for “a solution” based on these constraints. This is a good choice for a customer that wants a solution for a fixed cost.
- Optimization: Similar to above, the constraints need to be fully understood but rather than deliver a single suitable design they are paying for the optimum design. Typical optimizations might be:
- Minimize junction temperature
- Maintain junction temperature below a certain operating point and minimize the mass of the heat sink
- Maintain junction temperatures of inline components below a certain operating point and within a certain deltaT of each other and minimize the mass of the heat sinks.
Maybe a good way to distinguish between a design study and optimization would be to think about the words “Maintain” and “Minimize”. If you are minimizing or maximizing a parameter you are probably doing an optimization. Even then, if the design space isn’t properly positioned you may come up with a best design given the space but not necessarily an optimum.
When it comes to CFD analysis it is a lot easier to claim, than actually perform, an optimization. Typically an optimization would require, or at least benefit greatly from, a third party design-of-experiments software, not to mention the skills required to use that software. There is an exception with our tools, FloTHERM and FloVENT, which comes standard with all of this functionality. When I look back over the life of FloTHERM and FloVENT, and the road map it has taken, I can’t help but appreciate all of the visionaries and developers involved. The first step was to develop a tool with a robust solver and scheme that would lend itself to automation. The second key concept is the notion of SmartParts (i.e. Fans, Heat Sinks, etc) that provided the user all of the design knobs to vary a design. After this foundation had been laid and proven successful all of the power has been incrementally released to the community:
- Command Center: An interface that allows the user easy access to turn the knobs and collect the necessary analysis results to drive the design
- Design Of Experiments (DOE): Allows the user to study a design space using sophisticated mathematics (behind the scenes) to best study the space while limiting the number of analyses
- Sequential Optimization: Allows the user to pinpoint the optimum operating point based on the user’s criteria and made more effective by the DOE.
- Response Surface Optimization (RSO): Like DOE, this brings sophisticated mathematics (behind the scenes) directly to the thermal designer in Command Center. With this, the user can mathematically predict where the optimum operating point would be based on the DOE study. It also shows the sensitivity to design parameters as pointed out by Kelly Cordell-Morris in this blog. Without this functionality it would be nearly impossible to study the effect of changing one variable in an analysis that is varying multiple parameters.
As a politician might say, “Let’s be clear…” what we mean when we say optimization.